LiveVerified by Reviewer

No Human In The Loop

Find it. Fix it. Prove it.

Three autonomous agents scan your repo, patch what's broken inside a network-disabled sandbox, and run your real test suite to prove the fix holds.

The Pipeline

One handoff, three agents, zero waiting.

Detects

Watcher

Scans every file with Semgrep static analysis and GPT-5.6 semantic review, ranked by severity and confidence.

Patches

Codex Fixer

Writes the fix inside an isolated, network-disabled Docker sandbox, then validates the diff before it touches your repo.

Verifies

Reviewer

Re-runs your test suite against the patch and writes a plain-English explanation of what changed.

One real run

Not a mock. An actual fix.

This is one of the bugs seeded into AutoFix Swarm's own eval suite — a shipping threshold that quietly excluded the exact order total it was supposed to reward.

Watcher
Flagged the comparison at 92% confidence — the docstring promises “at or above,” the code only checked “above.”
Codex Fixer
Changed one operator inside a sandboxed clone of the repo. No network access, no other files touched.
Reviewer
Reran the shipping test suite against the patch — tests passed.
src/autofix_seed/shipping.pyfixed
  4  def shipping_cost(subtotal: float, free_shipping_threshold: float = 50.0) -> float:  5      """Return zero at or above the free-shipping threshold."""- 6      if subtotal > free_shipping_threshold:+ 6      if subtotal >= free_shipping_threshold:  7          return 0.0  8      return 8.99

Try it live

Watch the swarm work on real code.

Every run on the dashboard is a real Watcher → Codex Fixer → Reviewer pass, logged end to end.